Information Technology Reference
In-Depth Information
Each test sequence is encoded by IPPP prediction structure with the GOP length of
30. RDO is enabled both in mode decision and motion estimation. All test sequences
used are in 4:2:0 format, 150 Frames,
in Eq.(11) is set 0.5,
σ
in Eq.(13) is set 0.5.
The test platform is JM10.1 [13]. Like JVT-G012, the bit rates, which are generated
by encoding the test sequences with the fixed QPs of 28, 32, 36, and 40, are the target
bit rates for an encoder with the proposed rate control scheme. Table 1 illustrates
performance comparison between JVT-G012 and the proposed method. Experiments
have been carried out on PC with the Intel Core 3.0GHz CPU and 3.25GB RAM.
ζ
(a) RD performance for moniter
BDPSNR=0.541 dB
(b) RD performance for saleman
BDPSNR=0.248 dB
(d) RD performance for mumdaughter
BDPSNR=0.490 dB
Fig. 2. R-D curve comparison between the proposed method and JVT-G012
(c) RD performance for akiyo
BDPSNR= 0.361 dB
In the table 1, Rate error=(Actual Rate-Target Rate)/ Target Rate. It is clear that
JVT-G012 has a rate error range from 0.426% to 6.195%, while the proposed method
ranges from 3.254% to 7.686%, the proposed method outperforms JVT-G012. They
have similar rate error while the PSNR of the proposed method is improved up to
0.953 dB compared with JVT-G012, and the minimum improvement is 0.142dB. In
Table 1, it is seen that the proposed method achieves higher PSNR with negligible
and increments in bit rate. In order to compare coding efficiency, rate-distortion
Search WWH ::




Custom Search